pzhdfy commented on code in PR #68988:
URL: https://github.com/apache/doris/pull/68988#discussion_r4258231619
##########
fe/fe-core/src/main/java/org/apache/doris/datasource/paimon/source/PaimonScanNode.java:
##########
@@ -1156,6 +1266,13 @@ protected List<String> getDeleteFiles(TFileRangeDesc
rangeDesc) {
@Override
public List<Split> getSplits(int numBackends) throws UserException {
+ // Primary-key vector (ANN) search is planned entirely by Paimon's
+ // BatchVectorSearchBuilder: one BucketVectorSearchSplit per
(partition,
+ // bucket), serialized and carried to BE, which runs the whole search
+ // (recall + refine + row lookup + DV) in a single paimon-rust call.
+ if (isVectorSearch()) {
+ return getVectorSearchSplits();
Review Comment:
Agreed — stripping the cast chain searched the original vectors while the
query meant the casted ones, changing both the ranking and the reported
distance. The lossy ARRAY<FLOAT> -> ARRAY<INT> -> ARRAY<FLOAT> round trip is
not simplified away, so it was reachable.
▎
▎ Fixed in 4790f5a: the rewrite no longer strips casts at all; it accepts a
direct indexed column slot only. The lossy-array-cast negative rewrite case is
added to PushDownVectorTopNIntoPaimonScanTest.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
